Loadpull Compression point impedance using xdb analysis in Harmonic Balance

Hi SkkyLee,

I'm just curious, but what does the Pout vs. Pin plot look like if you were to pick a purely real value of 10 for the load impedance?

David Webster

Dear SkkyLee,

If you still have an interest, I took some time to study your data and re-analyze your transient simulation simulation results. I have attached the study I performed and included its summary page as Figure 1 to save you or anyone with an interest some time. Let me know if you have any questions or comments.

Shawn

Figure 1

Page 8 of Document "1db_compression_study_120823v1p0.pdf"

1db_compression_study_120823v1p0.pdf

Dear Shawn,

Thank you very much for your help! I have read your document carefully, last testbench to generate the results above has been updated, so I rebuilt the testbench to simulate, and the context below is based on the new rebulit testbench

• I don't know whether my understand is right that from your document maybe you think the 30.416dBm result was generated by the transient analysis? But I don't perform transient analysis, the result 30.416dBm OP1dB was gotten when I perform hb analysis's sweep function(as screenshot shown below), so I want to confirm that when perform pss/hb simulation's sweep function (for my instance, pin sweep from -20 to 30dBm), a set of transient analyses are executed?(From my teacher's point of view, when the snp file is right, the transient analysis is the most precise simulation, even though its efficiency is low), so does it mean that the sweep function's result (i.e. 30.416dBm) is more trustworthy compared to xdb simulation's result?

• From the suggestions of your document, I think the reason maybe:

1. the pss/hb analysis's result is fit to the small-signal transfer function gain 1.05, and the transient simulation analysis(pss/hb sweep power result?) is fit to the small-signal transfer function gain 1.00 (But I don't know how to verify this assumption, because I don't understand what the small-signal transfer function means exactly, does it mean the Power Amplifier's gain when the input power is very small?(Also my teacher's advice, he says the small-signal conception which used in the sp/ac analysis, the power level is always below -50dBm, but my PA simulation, the minimum input power is -20dBm, I don't know can this power level be called "small-signal"). And I don't know why the slope value is set as 1.05 the 2nd time(as the screenshot shown below), because of the "least squares algorithm"? Maybe I need to learn more knowledge...)

1. The pss/hb simulation the errpreset is moderate, and the transient simulation the errpreset is conservative(As screenshots below, I have set the accuracy settings as 'conservative', but the result remains nearly the same, there exist huge difference still...)

• About the gain peak, the power gain curve of this power amplifier shows that the gain peak exists definitely (as the screenshot shown below) which is fit to your document. But during this simulation process some new problems arisen:

1. The power gain is related to the impedance of inputport when using power gain(Direct plot) to plot the power gain, as the screenshot shown below, when I plot the power gain without finishing input matching between inputport and PA, the power gain read from the power gain curve(around 26dB) is not fit to the compression curve(around 10 to 11dB)

1.  When I perform lssp and set the inputport conjugate match with the input impedance of PA(as screenshot shown below the impedance of inputport set as 1.6+j*23), the power gain read from the power gain curve is fit to the compression curve(both around 20 to 22dB as screenshot shown below). So I think maybe this is the reason for the mismatch between the results of xdb and sweep compression point, so I want to perform xdb under this input conjugate matching condition, but this time the log file report "fatal error: assertion failed", this error I confronted when I perform lssp+sweep frequency, the solution is to set the the frequencies being swept the integer multiple of the fundamental frequency, which don't fit this time obviously which I perform loadpull+compression(xdb plot compression point loadpull contour). And I set the inputport impedance back to 50Ω, and this time the error disappears, so whether the inputport impedance is too small (1.6+j*23) and make the simulation don't converge?

The simulation above is performed in a short time (I am a little busy this month, sorry), so maybe exist mistakes or some important aspects I don't think about or understand...

Thank you very much!

This is the interesting phenomenon, when the real part impedance is bigger than the optimum load's real part, the xdb loadpull contour is fit to compression point sweep result very well, when the load impedance is 10Ω(it's bigger than the real part of the optimum load which is about 7Ω), the OP1dB is 30.39dBm, which fit to the xdb contour result.

Dear SkkyLee,

SkkyLee said:

But I don't perform transient analysis, the result 30.416dBm OP1dB was gotten when I perform hb analysis's sweep function(as screenshot shown below), so I want to confirm that when perform pss/hb simulation's sweep function (for my instance, pin sweep from -20 to 30dBm), a set of transient analyses are executed?

I now understand that the power transfer function plot is a result of an HB sweep of the input power and not a set of transient analyses. Thank you for this clarification SkkyLee.

SkkyLee said:

(But I don't know how to verify this assumption, because I don't understand what the small-signal transfer function means exactly, does it mean the Power Amplifier's gain when the input power is very small?(Also my teacher's advice, he says the small-signal conception which used in the sp/ac analysis, the power level is always below -50dBm, but my PA simulation, the minimum input power is -20dBm, I don't know can this power level be called "small-signal"). And I don't know why the slope value is set as 1.05 the 2nd time(as the screenshot shown below), because of the "least squares algorithm"?

To determine the X dB compression point, one must know the small-signal gain of the amplifier. The small-signal gain is the gain when the output shows essentially no harmonic distortion. The actual input power range that can be used to measure the small-signal gain is totally dependent on the gain characteristics of the amplifier under study. There is not a single range of input power levels that can be used to reliably provide a good estimate of the small-signal gain. In your plot of input to output power, I found that the slope and intercept used to compute the 1 dB compression point has a slope of 1.00 and an intercept of 12.166 dBm if one uses a range of input power levels between -20 dBm and -13 dBm . This produces your 1 dB compression point at an output power of 30.39 dBm.

However, if I use a smaller range of input powers to compute the linear fit (-20 dBm to -19 dBm), the slope and intercept are 1.05 and 12.984 dBm. When I used this curve to estimate the 1 dB compression point output power, I obtain 28.497 dBm. Hence, my hypothesis is that the small-signal gain of your amplifier in the sweep analysis is different than the small-signal gain of your amplifier in the Xdb analysis.

Unless you know the distortion characteristics of your amplifier output signal as a function of ihe input power. I think you want to use the smallest range of input power possible. I do not know how Cadence's tool chooses the range of input powers to use in its curve fit to your sweep. It may use the gain of "1.0" you provided in the form:

But, perhaps, you might try reducing the input power sweep to start at something much less than -20 dBm to see if the 1 dB  compression point output power is changed from 30.39 dBm.

SkkyLee said:

This is the interesting phenomenon, when the real part impedance is bigger than the optimum load's real part, the xdb loadpull contour is fit to compression point sweep result very well, when the load impedance is 10Ω(it's bigger than the real part of the optimum load which is about 7Ω), the OP1dB is 30.39dBm, which fit to the xdb contour

One of the reasons there is a discrepancy between your two values of 1 dB compresion point output power is the peaking in the power gain curve. This makes the 1 dB compression point output power very sensitive to the small-signal gain used in its estimate. When you change the load impedance from its matched value to use a real part of 10 ohms, it is likely the peaking in the power gain curve is no longer present. Intuitively, with the larger output impedance real part, my guess is the output current of your amplifier is less and hence it does not show the non-linear peaking in gain with input power. Therefore, if the output impedance changes the relationship between input and output power to eliminate the peak, I would expect the 1 dB compression power to be far less sensitive to the small-signal gain value and , as a result, show consistent results in your Xdb and HB sweep analyses.

Shawn